You are on page 1of 54

(II)

DSP (II)
DSP: Digital Signal Processing
Digital Signal Processor


-
2013.9.9 2011.12 .18
2013/9/8
2013/9/8

MMVCLAB
MMVCLAB

1.
2.DSP(I)
3.DSP
4.
DSP II
2013/9/8
2013/9/8

MMVCLAB
MMVCLAB

1.
1)
2)
3)
4)DP(Data Processing) VS. DSP
5) Why DSP ?

2013/9/8
2013/9/8

MMVCLAB
MMVCLAB

1.

1.
1): --
Examples:
A radio signal carries modulated music and speech.
A heartbeat signal (ECG) contains information about the
health of a persons heart.
Speech is an acoustic signal that humans use to transmit
information to each other.
A digitized, coded and modulated speech signal carries
speech signals.(PCM,Pulse Code Modulation)
----

2013/9/8
2013/9/8

MMVCLAB
MMVCLAB

----

Speech Recognition

Speaker Recognition

Language
Recognition

2013/9/8
2013/9/8

MMVCLAB
MMVCLAB

1.

----

2013/9/8
2013/9/8

MMVCLAB
MMVCLAB

1.

----

2013/9/8
2013/9/8

MMVCLAB
MMVCLAB

1.

2013/9/8
2013/9/8

MMVCLAB
MMVCLAB

1.

()

2013/9/8
2013/9/8

MMVCLAB
MMVCLAB

1.

2):
.
:,,
:

(I):
---()
(II):
----()
2013/9/8
2013/9/8

10

MMVCLAB
MMVCLAB

1.
xa (t )

A/D

3)
x(n)

y (n)

D/A

ya (t )

FFT,FIR,IIR

DSP

CPU
DSP

y=a*x+b---MAC,Multiply with ACcumulation 70%

2013/9/8
2013/9/8

11

MMVCLAB
MMVCLAB

1.

2013/9/8
2013/9/8

12

MMVCLAB
MMVCLAB

1.
DSP Board

2013/9/8
2013/9/8

13

MMVCLAB
MMVCLAB

Speech communication in human and in machine

Speech
Cognition

Planning

Language/
Knowledge

Motor
Command

Articulatory Movement

Speech Perception

Signal
Processing

2013/9/8
2013/9/8

14

MMVCLAB
MMVCLAB

1.

4)DP(Data Processing) VS. DSP


DP is the arithmetic processing of stored integer
numerical quantities (accounts, salary spread sheets and so
on). Fast processing of data is desirable but not essential.
DSP is concerned with the arithmetic processing of
numerical representations of real world analogue quantities.
Real time performance is necessary, such that processed
outputs are produced as fast as input data is available.
The key difference is :
zDSP - REAL TIME ARITHMETIC
zDP - NON REAL TIME ARITHMETIC
2013/9/8
2013/9/8

15

MMVCLAB
MMVCLAB

1.

All microprocessors can perform both tasks; however, it is difficult


(expensive) to make a device that is optimized for both. Traditional
microprocessors, such as the Pentium, primarily directed at data
manipulation. Similarly, DSPs are designed to perform the
mathematical calculations needed in Digital Signal Processing.

2013/9/8
2013/9/8

16

MMVCLAB
MMVCLAB

1.
5) Why DSP ?

Analog Systems
IN

vs.
OUT

Digital Systems
IN
A/D

2
+2
=4

OUT
D/A

- translate analog (e.g. filter) design into digital


- expand functionality/flexibility/
(e.g. analog speech recognition? analog.?)
2013/9/8
2013/9/8

17

MMVCLAB
MMVCLAB

1.

DSP in present-day systems?

Example: Telephone Line Modems


voice-band modems : up to 56kbits/sec in 0..4kHz
band
ADSL modems : up to 8Mbits/sec in 30kHz1MHz
band
(3.55km)
VDSL modems : up to 52Mbits/sec in
10MHz band
X 1000
(0.31.5km)
xDSL communication impairments:
channel attenuation/distortion, echo, cross-talk,...
2013/9/8
2013/9/8

18

MMVCLAB
MMVCLAB

1.

DSP in present-day systems?

Enabling Technology is
ASP
DSP-I
Signal Processing
DSP-II
1G-SP: analog filters
2G-SP: digital filters, FFTs, etc.
3G-SP: full of mathematics, linear algebra,
statistics, etc...
VLSI
etc...
2013/9/8
2013/9/8

19

MMVCLAB
MMVCLAB

2.DSP(I)
z
z
zFFT
z
zIIR
zFIR

2013/9/8
2013/9/8

20

MMVCLAB
MMVCLAB

2.DSP(I)

2.DSP(I)

x(t)

x*(t)=x(t)s(t)

s(t)

x(t)

x*(t)=x(t)s(t)
s(t)

t
T
2013/9/8
2013/9/8

21

MMVCLAB
MMVCLAB

2.DSP(I)

1
X * ( j) =
X [ j ( ns )]

T n =
|X*(j)|

|X(j)|

s >2h
-h

1/T

-h

2s

|X*(j)| s/2

s <2h 1/T
s

2s

s>=2h;s<2h
2013/9/8
2013/9/8

22

MMVCLAB
MMVCLAB

2.DSP(I)

:
(1);(2),
;(3),,
; (4)
DTFT, ,

Z;Z
;Z;
Z;Z;
.
(DFT)
2013/9/8
2013/9/8

23

MMVCLAB
MMVCLAB

2.DSP(I)

0 =

2
2
, 0 ==
NT
N

WN

=e

2
N

(,,)

N 1
~
nk
~
X ( k ) = x ( n )WN

k =0

1
~
x ( n) =
N

N 1

~
kn
X
(
k
)
W

N
k =0

2013/9/8
2013/9/8

24

MMVCLAB
MMVCLAB

2.DSP(I)
FFT

2DIT-FFT(FFT) Decimation In Time


2DIF-FFT(FFT) Decimation In Frequency
FFT

2013/9/8
2013/9/8

25

MMVCLAB
MMVCLAB

2.DSP(I)

N=

2M

X(0)

x(0)
x(4)

W N0 / 4
-1

x(2)
x(6)

W N0 / 2
W N1 / 2

0
N /4

X(1)

-1

-1

X(2)

-1

X(3)

x(1)
x(5)

W N1

W N0 / 4
-1

x(3)
x(7)

0
N /4

2013/9/8
2013/9/8

W N0

W N0 / 2
W

-1

1
N /2

W N2
-1

W
-1
26

3
N

-1

X(4)

-1

X(5)

-1
-1

X(6)
X(7)

MMVCLAB
MMVCLAB

2.DSP(I)

| H ( e j ) |

H ( e j ) = h ( n ) e j n

| H (e j ) |=| H (e j ) |
| H (e j ) |=| H (e j ( + 2 ) ) |

| H ( e j ) |

n =0

fs / 2

= T = 2f / f s

| H ( e j ) |

| H (e ) |

2013/9/8
2013/9/8

27

MMVCLAB
MMVCLAB

2.DSP(I)
x(n)

2013/9/8
2013/9/8

a1

z 1

a2

b1
b2

aM

aN

z 1

y(n)
M

H ( z) =

bM

br z

r =0
N

1 ak z
k =1

N M,

II()

28

MMVCLAB
MMVCLAB

2.DSP(I)

N 1

y (n ) = h(k ) x(n k )
k =0

x(n)
h(0)
N 1

z 1 z
h(1)

H ( z ) = h( k ) z

z 1 z

h(2)

( N 1)
z
1

h(N 1)

y(n)

k =0

N,
2013/9/8
2013/9/8

29

MMVCLAB
MMVCLAB

2.DSP(I)
IIR

FIR

FIR

2013/9/8
2013/9/8

30

MMVCLAB
MMVCLAB

3.DSP
z
z
zIC

2013/9/8
2013/9/8

31

MMVCLAB
MMVCLAB

3.DSP

3.DSP

1) :(PSTN)->->->,
,
(,,); ()

2) CD,MP3,VIDEO,DIGITAL VIDEO
CAMARA
3) (DAB),HDTV
4) ,,
2013/9/8
2013/9/8

32

MMVCLAB
MMVCLAB

3.DSP
Many DSP applications are converging to single general purpose application
programmable DSP systems. For example the emerging personal
communicator technology in the not too distant future will have the following
features which rely on a number of DSP algorithms and technology for
implementation:

2013/9/8
2013/9/8

33

MMVCLAB
MMVCLAB

3.DSP

1) Source coding (speech coding, Image


Coding)
Real World Sampling Frequency
CD Audio: 44100 Hz
Professional Digital Audio: 48000 Hz
Telephone Speech: 8000 Hz
Teleconferencing: 16000 Hz
Biomedical: 1000 Hz
Control Systems: 100 Hz
2013/9/8
2013/9/8

34

MMVCLAB
MMVCLAB

3.DSP
Real World Sample Bits
CD Audio: 16 bits
Professional Digital Audio: 24 bits
Telephone Speech: 8 - 13 bits
Teleconferencing: 12 - 16 bits
Biomedical: 16 - 24 bits
Control Systems: 4 - 10 bits

2013/9/8
2013/9/8

35

MMVCLAB
MMVCLAB

3.DSP
Real World Sample Bit Rates = Sample Bits X
Sampling Frequency
CD Audio: 705.6kbps
Professional Digital Audio: 1.152Mbps
Telephone Speech: 64~104kbps
Teleconferencing: 192~256kbps
Biomedical: 16~24kbps
Control Systems: 400 1000bps

2013/9/8
2013/9/8

36

MMVCLAB
MMVCLAB

3.DSP
2)---
Signal separation/Blind Channel Equalization:
,Cocktail-Party
x1(t) = 11s1(t) +12s2 (t) +13s3 (t )

0 .5

x2 (t) = 21s1(t) +22s2 (t) +23s3 (t )

-0 . 5
0

0 .5

1.5

2 .5

3 .5

4 .5

x3 (t ) = 31s1(t ) +32s2 (t ) +33s3 (t )

5
x 10

0 .5
0
-0 . 5
0

0 .5

1.5

2 .5

3 .5

4 .5

10

5
x 10

0
-5
0

-10

0 .5

1.5

2 .5

3 .5

4 .5

5
x 10

-20
0

0.5

1.5

2.5

3.5

4.5

10

5
x 10

0
-2
0

0.5

1.5

2 .5

3 .5

4.5

-10
0

0.5

1.5

2.5

3.5

4.5

5
x 10

5
x 10

-5

0.5

1.5

2 .5

3 .5

4.5

5
x 10

-5
0

0.5

1.5

2.5

3.5

4.5

5
x 10

-5
0

2013/9/8
2013/9/8

37

0.5

1.5

2 .5

3 .5

4.5

MMVCLAB
MMVCLAB
x 10

3.DSP
Cocktail Party Effect:Hear More than one voice

Four psychoacoustically heuristic regularities related to acoustic events:


1. common onset and offset,
2. gradualness of change,
3. harmonicity, and
4.
changes occurring in the acoustic event
2013/9/8
2013/9/8

38

MMVCLAB
MMVCLAB

3.DSP

2013/9/8
2013/9/8

39

MMVCLAB
MMVCLAB

3.DSP
3)
Background noise:Robust

ETSI ES 202 050


V1.1.1 (2002-10)

2013/9/8
2013/9/8

40

MMVCLAB
MMVCLAB

3.DSP
IC
Architecture of the Digital Signal Processor

2013/9/8
2013/9/8

41

MMVCLAB
MMVCLAB

3.DSP

The Von Neumann architecture uses a single memory to hold both


data and instructions. In comparison, the Harvard architecture uses
separate memories for data and instructions, providing higher speed.
The Super Harvard Architecture improves upon the Harvard design
by adding an instruction cache and a dedicated I/O controller.
2013/9/8
2013/9/8

42

MMVCLAB
MMVCLAB

3.DSP

Typical DSP architecture.


Digital Signal Processors are
designed to implement tasks
in parallel.

2013/9/8
2013/9/8

43

MMVCLAB
MMVCLAB

3.DSP
DSP Speed

fixed point systems are often quoted in MIPS(Million


Integer operations Per Ssecond).
floating point devices can be specified in MFLOPS
(Million FLOating point operations Per Ssecond).
MMAC(Million Multiplication-ACcumulation)

2013/9/8
2013/9/8

44

MMVCLAB
MMVCLAB

3.DSP
Example:SHARC DSPS

The throughput of a particular DSP algorithm can be found by dividing the


clock rate by the required number of clock cycles per sample. This illustration
shows the range of the throughput for four common algorithms, executed on a
SHARC DSP at a clock speed of 40 MHz.

2013/9/8
2013/9/8

45

MMVCLAB
MMVCLAB

3.DSP
Example:-Blackfin DSPs

Benchmarks for three algorithms

Algorithm Type
256-pt Complex FFT
FIR Filter (per Tap)
IIR Filter (per Biquad)

2013/9/8
2013/9/8

Time

Cycles

0.0106 ms 3,176
13.33 ns
4
20 ns
6

46

MMVCLAB
MMVCLAB

2013/9/8
2013/9/8

47

MMVCLAB
MMVCLAB

2013/9/8
2013/9/8

48

MMVCLAB
MMVCLAB

3.DSP

2013/9/8
2013/9/8

49

MMVCLAB
MMVCLAB

4.DSP II

1),Adaptive Filtering
2)
Short Time Fourier Transform,Wavelet Transform
3)
Modern Spectrum Estimation
4)
Multi-rate Processing of Digital Signal
5)()
Non-Gaussian Signal Processing
6),
Classification and Recognition
(VQ,Neural network,HMM,)
2013/9/8
2013/9/8

50

MMVCLAB
MMVCLAB

4.DSP II

DSP II
1)(Weiner Filtering,
Kalman Filtering, LMS Adaptive Filtering,RLS
Adaptive Filtering,Linear Predictive Error
Filtering)
2)(,
)
3)()
4)(,,
,)
2013/9/8
2013/9/8

51

MMVCLAB
MMVCLAB

4.DSP II

A. Oppenheim & R. Schafer


DSP-I
`Digital Signal Processing (Prentice Hall, 1977)

1999

1995
DSP-II
Simon Haykin
`Adaptive Filter Theory (Prentice Hall, 2002)
Ali H. Saved
Fundamentals of Adptive Filtering
(IEEE Press,2003)
2013/9/8
2013/9/8

52

MMVCLAB
MMVCLAB


1)
2)
3)

2013/9/8
2013/9/8

53

MMVCLAB
MMVCLAB


(
)

lrdai@ustc.edu.cn
6BBS
2013/9/8
2013/9/8

54

MMVCLAB
MMVCLAB

You might also like